This message was deleted.

Whatever problem you are hoping to solve here, I cannot imagine the juice would be anywhere worth the squeeze. So, what problem are you trying to solve? 🙂

What a would think would be a fairly common problem: your nodes have a hardware requirement (e.g. GPU nodes) and you're trying to aggregate the them based on dynamic pricing - with some mix of your own (on-prem) and cloud provided. And it seems that most solutions our here focus on multiple clusters across multiple clouds, but a single cluster across multiple clouds seems easier to manage. To some extend this may only be easier because our nodes don't need to talk to eachother (so networking is less complicated).

I'd imagine a custom RKE2 cluster would work with multiple cloud providers if you aren't worried about cloud provider integration, since you're just running the registration command on the nodes themselves. I have not tried it though so I could be wrong.

I’m left wondering if this is even a job for k8s. 🤷

Actually this was my conclusion as well. Since I don't even need services that talk to eachother or anything like that, most of the benefits of K8s aren't really there. I'm currently just trying just to use an AWS feature called ECS Anywhere, which is supposed to allow you to just add worker instances to an existing ECS cluster.

My mind aligns with yours on this. You might even ask yourselves “Is containerizing this workload the right move?” 😉